An energy management system and method for energy storage batteries

Abstract

This application proposes an energy management system and method for energy storage batteries. The system includes: a data acquisition and processing module; a battery state estimation module connected to the data acquisition and processing module; an adaptive scenario-based energy scheduling module connected to both the data acquisition and processing module and the battery state estimation module; and an execution module connected to the adaptive scenario-based energy scheduling module. The adaptive scenario-based energy scheduling module includes: a scenario identification unit, a strategy generation unit, and an execution feedback unit. The technical solution proposed in this application, by constructing a closed-loop intelligent scheduling architecture of "perception-decision-feedback," solves the problems of poor scenario adaptability and rigid strategies in traditional systems, significantly improving photovoltaic absorption rate, user economic benefits, and battery lifespan, and enhancing the system's adaptability to dynamic environments and long-term changes.

Description

Technical Field

[0001] This application relates to the field of battery management technology, and in particular to an energy management system and method for energy storage batteries. Background Technology

[0002] A battery management system (BMS) is the core control unit used to monitor, manage, and protect energy storage battery packs, playing a crucial role as the "brain" of the energy storage system. It achieves comprehensive battery status awareness by real-time acquisition of key parameters such as voltage, current, and temperature of individual cells or modules within the battery pack, and accurately calculating core indicators such as state of charge (SOC) and state of health (SOH). Based on this, the BMS maintains consistency between cells through active balancing technology, optimizes charging and discharging strategies to extend battery life, and rapidly executes protective actions to disconnect circuits and ensure safety in the event of abnormal conditions such as overvoltage, undervoltage, overcurrent, or overtemperature. As a bridge connecting batteries and application scenarios, a high-performance BMS is the cornerstone for ensuring the safe, stable, and efficient operation of energy storage power stations, electric vehicles, and other systems.

[0003] Existing technologies typically lack accurate multi-scenario identification and prioritization capabilities. Strategy generation relies on fixed rules rather than adaptive learning models, failing to adjust charging and discharging strategies based on dynamic changes such as photovoltaic output fluctuations and battery degradation. Furthermore, the lack of a closed-loop feedback mechanism hinders continuous optimization of strategy effectiveness, leading to problems such as low photovoltaic grid integration rates, insufficient economic returns, shortened battery life, or delayed emergency backup power response. Therefore, there is an urgent need to propose an energy management solution for energy storage batteries that can accurately perceive complex scenarios, possess adaptive learning capabilities, and continuously optimize, in order to improve the overall operational efficiency of energy storage systems. Summary of the Invention

[0004] This application provides an energy storage battery energy management system and method to at least solve the technical problems of low photovoltaic absorption rate, insufficient economic benefits, damaged battery life and delayed emergency demand response caused by the poor adaptability of existing energy storage management systems, fixed strategies and lack of self-learning ability.

[0005] A first aspect of this application provides an energy management system for an energy storage battery, the system comprising:

[0006] The data acquisition and processing module is used to collect and fuse internal battery status data with external environmental data. The battery state estimation module is connected to the data acquisition and processing module and is used to receive the internal state data of the battery and estimate the battery's state of charge (SOC), state of health (SOH), and state of function (SOF). The adaptive scenario-based energy scheduling module is connected to the data acquisition and processing module and the battery state estimation module respectively, and is used to receive the external environment data and the battery's SOC, SOH, and SOF, and generate the optimal energy scheduling strategy. The execution module is connected to the adaptive scenario-based energy scheduling module and is used to execute the optimal energy scheduling strategy; The adaptive scenario-based energy scheduling module includes: The scene recognition unit is used to extract features and recognize patterns from the external environment data and battery status data to determine the current scene type and its priority. The strategy generation unit has a built-in improved reinforcement learning decision model, which is used to generate the optimal charge and discharge management strategy based on the scene and priority determined by the scene recognition unit and the output of the battery state estimation module. The execution feedback unit is used to monitor the execution effect of the execution module on the strategy and generate feedback samples based on the actual data after execution to optimize the decision model in the strategy generation unit.

[0007] Preferably, the scene recognition unit includes: The rule matching library pre-stores trigger conditions for multiple preset scenarios; The clustering analysis submodule is used to perform clustering analysis on real-time collected external environment data to identify dynamic sub-patterns in the data. A scene feature extraction unit is used to extract key features for scene determination from the external environment data and battery status data. The priority determination submodule is used to assign or dynamically adjust the priority of the identified current scene based on the matching results of the rule matching library, the output of the clustering analysis submodule, and the key features.

[0008] Furthermore, the preset scenarios include: peak-valley arbitrage scenario, photovoltaic consumption scenario, emergency backup power scenario, and grid auxiliary service scenario; The priority determination submodule is configured to assign the highest priority when an emergency backup power scenario is identified.

[0009] Preferably, the improved reinforcement learning decision model in the policy generation unit is an improved Q-learning model, and its Q-value update formula is:

[0010] In the formula, Q(s,a) is the Q value of performing action a in the current state s, α is the learning rate, R is the immediate reward, λ is the battery health cost coefficient, and Csoh is the battery health depletion cost caused by action a. γ is the scene weighting factor, and γ is the discount factor. It is the maximum Q value of all possible actions in the next state s′.

[0011] Preferably, the adaptive scenario-based energy scheduling module further includes: The AI ​​prediction unit is connected to the data acquisition and processing module and the strategy generation unit, respectively, and is used to predict future user electricity load, photovoltaic output and / or grid electricity price based on historical data and real-time data, and input the prediction results to the strategy generation unit.

[0012] Furthermore, the AI ​​prediction unit incorporates an improved LSTM prediction model, and the forget gate calculation formula for the LSTM prediction model is as follows:

[0013] In the formula, The output of the forget gate at time t takes a value between 0 and 1, and σ is the Sigmoid activation function. To hide the state of history through the Forgotten Gate and current input The weight matrix, Let be the hidden state vector at time t-1. Let be the input feature vector at time t, and Ws be the weight matrix of the forget gate for the battery's SOC state. Let be the remaining battery charge at time t. This is the bias term for the forget gate.

[0014] Furthermore, the improved LSTM prediction model is trained using a weighted loss function, the formula for which the prime number loss function is calculated is:

[0015] In the formula, L is the total model loss, α is the weighting coefficient of the prediction accuracy term, and MAE is the mean absolute error. The calculation formula is as follows: Where: n is the total number of samples involved in the calculation, and i is the sample index, ranging from 1 to n, representing the i-th specific sample. for , The true value of the i-th sample, |...| is used to convert the difference between the predicted value and the true value into a non-negative number, avoiding the cancellation of positive and negative errors. For the predicted electricity load vector, Let be the actual electrical load vector, and β be the weighting coefficient of the SOC constraint term. For the predicted battery SOC value, This represents the user's desired battery SOC target value.

[0016] Preferably, the execution feedback unit includes: An action execution monitor is used to track and record the actual execution of the strategy by the execution module; The real-time data acquisition submodule is used to collect system status data at a preset frequency after the strategy is executed. The feedback sample generation unit is used to integrate the state data before the action is executed, the action information after the action is executed, and the state data after the action is executed into a standardized feedback sample. The model update triggering submodule is used to initiate the update of the decision model in the strategy generation unit based on preset conditions and the generated feedback samples.

[0017] Preferably, the system further includes: The equalization management module, connected to the battery state estimation module and the adaptive scenario-based energy scheduling module, is used to perform active or passive equalization between battery cells according to the battery state and the optimal energy scheduling strategy. The safety protection module is connected to the data acquisition and processing module and the adaptive scenario-based energy scheduling module, and is used to perform emergency protection actions based on the acquired fault signals. The communication and interaction module is connected to the battery state estimation module and the adaptive scenario-based energy scheduling module, and is used to realize information interaction between the system and users, the cloud and the power grid.

[0018] A second aspect of this application provides a method for energy management of a storage battery, including: Real-time acquisition of battery internal status data and external environmental data; Estimate the battery's SOC, SOH, and SOF based on the battery's internal state data; The collected data is identified to determine the current operating scenario and its priority; An improved reinforcement learning decision model is used to generate an optimal energy scheduling strategy based on the scenario, priority, and battery state. Execute the optimal energy scheduling strategy; The execution feedback unit is used to monitor the effect of the strategy execution and generate feedback samples for iterative optimization of the reinforcement learning decision model.

[0019] The technical solutions provided by the embodiments of this application have at least the following beneficial effects: This application proposes an energy management system and method for energy storage batteries. The system includes: a data acquisition and processing module for acquiring and fusing internal battery state data and external environmental data; a battery state estimation module connected to the data acquisition and processing module for receiving the internal battery state data and estimating the battery's state of charge (SOC), state of health (SOH), and state of function (SOF); an adaptive scenario-based energy scheduling module connected to both the data acquisition and processing module and the battery state estimation module for receiving the external environmental data and the battery's SOC, SOH, and SOF, and generating an optimal energy scheduling strategy; and an execution module connected to the adaptive scenario-based energy scheduling module. The adaptive scenario-based energy scheduling module is connected to execute the optimal energy scheduling strategy. The module includes: a scenario identification unit for extracting features and recognizing patterns from external environmental data and battery state data to determine the current scenario type and its priority; a strategy generation unit with a built-in improved reinforcement learning decision model for generating an optimal charge / discharge management strategy based on the scenario and priority determined by the scenario identification unit and the output of the battery state estimation module; and an execution feedback unit for monitoring the execution effect of the strategy by the execution module and generating feedback samples based on the actual data after execution to optimize the decision model in the strategy generation unit. The technical solution proposed in this application, by constructing a closed-loop intelligent scheduling architecture of "perception-decision-feedback," solves the problems of poor scenario adaptability and rigid strategies in traditional systems, significantly improving photovoltaic absorption rate, user economic benefits, and battery lifespan, and enhancing the system's adaptability to dynamic environments and long-term changes.

[0020] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0021] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a first structural diagram of an energy storage battery energy management system according to an embodiment of this application; Figure 2 This is a structural diagram of a data acquisition and processing module provided according to an embodiment of this application; Figure 3 This is a structural diagram of a battery state estimation module according to an embodiment of this application; Figure 4 This is a structural diagram of a scene recognition unit provided according to an embodiment of this application; Figure 5 This is a structural diagram of a strategy generation unit provided according to an embodiment of this application; Figure 6 This is a structural diagram of an adaptive scenario-based energy scheduling module provided according to an embodiment of this application; Figure 7 This is a structural diagram of an execution feedback unit provided according to an embodiment of this application; Figure 8 This is a second structural diagram of an energy storage battery energy management system according to an embodiment of this application; Figure 9 This is a structural diagram of a load balancer management module provided according to an embodiment of this application; Figure 10 This is a structural diagram of a security protection module provided according to an embodiment of this application; Figure 11 This is a structural diagram of a communication and interaction module provided according to an embodiment of this application; Figure Labels Data Acquisition and Processing Module 1; Battery State Estimation Module 2; Adaptive Scenario-Based Energy Scheduling Module 3; Execution Module 4; Balanced Management Module 5; Safety Protection Module 6; Communication and Interaction Module 7; Sensor Group 1-1; Data Preprocessing Unit 1-2; Time Series Data Buffer 1-3; Multi-Source Data Fusion Submodule 1-4; SOC Estimation Submodule 2-1; SOH Evaluation Submodule 2-2; SOF Calculation Submodule 2-3; State Fusion Output Unit 2-4; Scene Recognition Unit 3-1; Strategy Generation Unit 3-2; AI Prediction Unit 3-4; Execution Feedback Unit 3-3; Rule Matching Library 3-1-1; Cluster Analysis Submodule 3-1-2; Scene Feature Extraction Unit 3-1- 3. Priority determination submodule 3-1-4, multi-source data input layer 3-2-1, scene intelligent recognition module 3-2-2, improved Q-learning decision model 3-2-3 and execution optimization module 3-2-3, action execution monitor 3-3-1, real-time data acquisition submodule 3-3-2, feedback sample generation unit 3-3-3, model update triggering submodule 3-3-4, equilibrium strategy selection unit 5-1, active equilibrium circuit control submodule 5-2, passive equilibrium triggering submodule 5-3, equilibrium state monitoring unit 5-4, local communication interface 7-1, cloud data transmission unit 7-2, user interaction interface submodule 7-3, instruction parsing submodule 7-4. Detailed Implementation

[0022] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0023] This application proposes an energy management system and method for an energy storage battery. The system includes: a data acquisition and processing module for acquiring and fusing internal battery state data and external environmental data; a battery state estimation module connected to the data acquisition and processing module for receiving the internal battery state data and estimating the battery's state of charge (SOC), state of health (SOH), and state of function (SOF); an adaptive scenario-based energy scheduling module connected to both the data acquisition and processing module and the battery state estimation module for receiving the external environmental data and the battery's SOC, SOH, and SOF, and generating an optimal energy scheduling strategy; and an execution module connected to the adaptive scenario-based energy scheduling module. The adaptive scenario-based energy scheduling module is connected to execute the optimal energy scheduling strategy. The module includes: a scenario identification unit for extracting features and recognizing patterns from external environmental data and battery state data to determine the current scenario type and its priority; a strategy generation unit with a built-in improved reinforcement learning decision model for generating an optimal charge / discharge management strategy based on the scenario and priority determined by the scenario identification unit and the output of the battery state estimation module; and an execution feedback unit for monitoring the execution effect of the strategy by the execution module and generating feedback samples based on the actual data after execution to optimize the decision model in the strategy generation unit. The technical solution proposed in this application, by constructing a closed-loop intelligent scheduling architecture of "perception-decision-feedback," solves the problems of poor scenario adaptability and rigid strategies in traditional systems, significantly improving photovoltaic absorption rate, user economic benefits, and battery lifespan, and enhancing the system's adaptability to dynamic environments and long-term changes.

[0024] An energy management system and method for an energy storage battery according to an embodiment of this application are described below with reference to the accompanying drawings.

[0025] Example 1 Figure 1 This is a structural diagram of an energy storage battery energy management system according to an embodiment of this application, as shown below. Figure 1 As shown, the system includes: Data acquisition and processing module 1 is used to collect and fuse internal battery status data and external environmental data; The battery state estimation module 2 is connected to the data acquisition and processing module 1 and is used to receive the internal state data of the battery and estimate the battery's state of charge (SOC), state of health (SOH), and state of function (SOF). The adaptive scenario-based energy scheduling module 3 is connected to the data acquisition and processing module 1 and the battery state estimation module 2 respectively, and is used to receive the external environment data and the battery's SOC, SOH, and SOF, and generate the optimal energy scheduling strategy. Execution module 4 is connected to the adaptive scenario-based energy scheduling module 3 and is used to execute the optimal energy scheduling strategy; The adaptive scenario-based energy scheduling module includes: Scene recognition unit 3-1 is used to extract features and recognize patterns from the external environment data and battery status data to determine the current running scene type and its priority; The strategy generation unit 3-2 has a built-in improved reinforcement learning decision model, which is used to generate the optimal charge and discharge management strategy based on the scene and priority determined by the scene recognition unit and the output of the battery state estimation module. The execution feedback unit 3-3 is used to monitor the execution effect of the execution module on the strategy and generate feedback samples based on the actual data after execution to optimize the decision model in the strategy generation unit.

[0026] It should be noted that, as Figure 2 As shown, the data acquisition and processing module 1 is internally equipped with a sensor group 1-1, a data preprocessing unit 1-2, a time-series data buffer 1-3, and a multi-source data fusion submodule 1-4; Sensor group 1-1 collects data on individual battery cell voltage, total current, cell temperature, external grid electricity price signal, photovoltaic power output data, and real-time user electricity load. Data preprocessing units 1-2 perform filtering, noise reduction, and standardization transformation on the collected raw data; Time-series data caches 1-3 store historical monitoring data for the most recent 72 hours; Multi-source data fusion submodules 1-4 integrate battery status data and external environment data into a unified input stream, providing continuous and reliable data support for subsequent modules.

[0027] It should be noted that, as Figure 3 As shown, the battery state estimation module 2 is internally configured with a SOC estimation submodule 2-1, a SOH evaluation submodule 2-2, a SOF calculation submodule 2-3, and a state fusion output unit 2-4; SOC estimation submodule 2-1 uses an extended Kalman filter algorithm to process voltage and current data; SOH evaluation submodule 2-2 calculates the state of health based on the number of battery cycles and capacity decay rate; SOF calculation submodule 2-3 determines the maximum charge and discharge power by combining the current temperature and voltage values; state fusion output unit 2-4 integrates the results of the three submodules into comprehensive battery state parameters for use by the strategy generation unit.

[0028] In the embodiments disclosed herein, such as Figure 4 As shown, the scene recognition unit 3-1 includes: The rule matching library 3-1-1 has pre-stored trigger conditions for multiple preset scenarios; The clustering analysis submodule 3-1-2 is used to perform clustering analysis on real-time collected external environment data and identify dynamic sub-patterns in the data. Scene feature extraction unit 3-1-3 is used to extract key features for scene determination from the external environment data and battery status data; Priority determination submodule 3-1-4 is used to assign or dynamically adjust the priority of the identified current scene based on the matching results of the rule matching library, the output of the clustering analysis submodule, and the key features.

[0029] It should be noted that the preset scenarios include: peak-valley arbitrage scenario, photovoltaic consumption scenario, emergency backup power scenario, and grid auxiliary service scenario; The priority determination submodule 3-1-4 is configured to assign the highest priority when an emergency backup power scenario is identified.

[0030] It should be noted that the internal storage of the rule matching library 3-1-1 contains trigger conditions for scenarios such as peak-valley arbitrage, photovoltaic consumption, emergency backup power, and grid ancillary services. Among them, the trigger conditions for peak-valley arbitrage include the current electricity price being in the system's preset peak segment and the remaining battery power being less than 50%. The trigger conditions for photovoltaic consumption include the real-time photovoltaic output exceeding the user's real-time load and the remaining battery power being less than 80%. The trigger conditions for emergency backup power include receiving an emergency signal manually sent by the user or a grid outage alarm. The trigger conditions for grid ancillary services include receiving a frequency regulation or peak regulation instruction issued by the grid dispatch center.

[0031] The clustering analysis submodule 3-1-2 performs K-means clustering on the grid electricity price data, photovoltaic output curves, and user load fluctuation data within the last 15 minutes, dividing them into sub-scenario categories such as stable load, high photovoltaic output, and peak electricity price, providing supplementary reference for rule matching.

[0032] The scene feature extraction unit 3-1-3 extracts key features from multi-source input data, including extracting time period labels (morning peak, noon peak, evening peak, valley) from electricity price data, extracting power output level (high, medium, low) from photovoltaic data, extracting remaining power range (0-20%, 20%-50%, 50%-80%, 80%-100%) from battery data, and extracting load type identifiers from user data.

[0033] The priority determination submodule 3-1-4 determines the scenario priority based on the rule matching results and feature extraction data. For example, when both peak-valley arbitrage and photovoltaic consumption conditions are met, if the photovoltaic output level is high, the photovoltaic consumption scenario is given priority. If the difference between the current electricity price and the off-peak electricity price exceeds 0.3 yuan / kWh, the peak-valley arbitrage scenario is given priority. The emergency backup power scenario is directly assigned the highest priority without comparison with other scenarios.

[0034] In this embodiment of the disclosure, the improved reinforcement learning decision model in the policy generation unit 3-2 is an improved Q-learning model, and its Q-value update formula is:

[0035] In the formula, Q(s,a) is the Q value of taking action a in the current state s (reflecting the expected long-term return of the action), α is the learning rate (controlling the model update amplitude, ranging from 0 to 1), R is the immediate reward such as the electricity price difference benefit from peak-valley arbitrage, or the clean energy savings from photovoltaic consumption), λ is the battery health cost coefficient (adjusting the weight of health loss in the decision-making process, ranging from 0 to 0.5), and Csoh is the battery health loss cost caused by action a (calculated based on the SOH decay rate, in yuan / time). γ is the scenario weighting factor (1.2 for emergency backup power scenarios and 0.8 for peak-valley arbitrage, reflecting scenario priority), and γ is the discount factor (measuring the importance of future returns, with a value ranging from 0 to 1). It is the maximum Q value of all possible actions in the next state s′.

[0036] This module integrates battery health degradation costs and scenario priorities into the traditional Q-learning decision framework during the calculation process. This ensures that energy storage benefits are maximized while taking into account battery life and the urgency of the scenario, making it more in line with the actual operation requirements of energy storage systems.

[0037] It should be noted that, as Figure 5 As shown, the strategy generation unit 3-2 includes: a multi-source data input layer 3-2-1, a scene intelligent recognition module 3-2-2, an improved Q-learning decision model 3-2-3, and an execution optimization module 3-2-3.

[0038] Furthermore, such as Figure 6 As shown, the adaptive scenario-based energy scheduling module 3 further includes: AI prediction unit 3-4 is connected to the data acquisition and processing module 1 and the strategy generation unit 3-2 respectively. It is used to predict future user electricity load, photovoltaic output and / or grid electricity price based on historical data and real-time data, and input the prediction results to the strategy generation unit 3-2.

[0039] It should be noted that the AI ​​prediction unit incorporates an improved LSTM prediction model, and the forget gate calculation formula for the LSTM prediction model is as follows:

[0040] In the formula, The output of the forget gate at time t takes a value between 0 and 1, and σ is the Sigmoid activation function. To hide the state of history through the Forgotten Gate and current input The weight matrix, Let be the hidden state vector at time t-1. Let be the input feature vector at time t, and Ws be the weight matrix of the forget gate for the battery's SOC state. Let be the remaining battery charge at time t. This is the bias term for the forget gate.

[0041] The improved LSTM prediction model is trained using a weighted loss function, the formula for which is:

[0042] In the formula, L is the total model loss, α is the weighting coefficient of the prediction accuracy term, and MAE is the mean absolute error. The calculation formula is as follows: Where: n is the total number of samples involved in the calculation, and i is the sample index, ranging from 1 to n, representing the i-th specific sample. for , The true value of the i-th sample, |...| is used to convert the difference between the predicted value and the true value into a non-negative number, avoiding the cancellation of positive and negative errors. For the predicted electricity load vector, Let be the actual electrical load vector, and β be the weighting coefficient of the SOC constraint term. For the predicted battery SOC value, This represents the user's desired battery SOC target value.

[0043] It should be noted that the AI ​​prediction unit 3-4 internally includes a multi-dimensional feature processing module (responsible for collecting data such as user electricity load, battery SOC, ambient temperature, weather type, and grid electricity price and performing standardized preprocessing), an improved LSTM model (introducing a battery SOC state influencing factor into the core forget gate mechanism to enhance the ability to capture the dynamic correlation between battery state and electricity demand in energy storage scenarios), a weighted loss function training module (achieving multi-objective optimization by combining prediction accuracy terms and SOC constraint terms, taking into account both load prediction accuracy and user expectations for battery state), and a real-time feedback update module (continuously iterating and optimizing model weights to improve scenario adaptability by utilizing actual battery state data after execution and user / grid feedback information).

[0044] The AI ​​prediction unit 3-4 first collects multi-dimensional features, including hourly electricity load data from users over the past 30 days, real-time battery SOC values, ambient temperature, weather type (sunny / rainy / cloudy), and peak / valley electricity price periods. These features are standardized and used as model input. The model employs an improved Long Short-Term Memory (LSTM) network structure. Considering the specific characteristics of energy storage scenarios, the influence factor of battery SOC state is introduced into the forget gate of the traditional LSTM, enabling the model to better capture the correlation between battery state and electricity demand. During the training phase, a sliding window method is used to generate samples (window size is 24 hours, predicting the load for the next hour). The loss function is designed as a weighted combination of prediction accuracy and battery SOC constraints, ensuring both the accuracy of load prediction and taking into account users' expectations of battery state (such as maintaining a high SOC during nighttime backup power). After model training, it can output hourly electricity load predictions for the next 24 hours. Simultaneously, by combining real-time collected actual load data, the model weights are periodically updated using gradient descent to continuously optimize the prediction performance. In addition, when users manually adjust energy storage strategies (such as temporarily adding electrical equipment), the model will use this feedback information as a new training sample to further improve scenario adaptability.

[0045] In the embodiments disclosed herein, such as Figure 7 As shown, the execution feedback unit 3-3 includes: Action execution monitor 3-3-1 is used to track and record the actual execution of the strategy by the execution module; The real-time data acquisition submodule 3-3-2 is used to collect system status data at a preset frequency after the strategy is executed. The feedback sample generation unit 3-3-3 is used to integrate the state data before the action is executed, the action information after the action is executed, and the state data after the action is executed into a standardized feedback sample; The model update triggering submodule 3-3-4 is used to initiate the update of the decision model in the strategy generation unit based on the generated feedback samples according to preset conditions.

[0046] It should be noted that the action execution monitor 3-3-1 tracks the actual execution of all actions output by the strategy generation unit, including the switching status of the charging and discharging circuit, the current change value of the equalization circuit, the deviation between the output voltage of the power converter and the preset value, and records the start time, duration, and difference between the actual power level and the strategy output level of the action execution.

[0047] The real-time data acquisition submodule 3-3-2 collects data at different frequencies after the action is executed. It collects the battery cell voltage, total current, and cell temperature every 30 seconds, the grid electricity price update value, the actual value of photovoltaic output, and the real-time value of user load every 5 minutes, and the change in battery remaining power and the health status decay value every hour.

[0048] The feedback sample generation unit 3-3-3 integrates the state data before the action is executed, the action information after the action is executed, and the state data after the action is executed into standardized samples. For example, the sample contains 12 fields such as the remaining battery power range before the action is executed, the electricity price period, the photovoltaic output level, the charging and discharging power level executed, the remaining battery power after the action is executed, and the actual arbitrage profit generated. The samples are stored in the local database in chronological order.

[0049] The model update triggering submodule 3-3-4 determines whether to start the model update based on the number of samples and the deviation. When the cumulative number of samples in the local database reaches 1000, the update process is automatically started. When the deviation between the actual result and the prediction result after 5 consecutive actions exceeds 15%, parameter adjustment is immediately triggered. After the update process is started, the sample set is divided into a training set and a validation set for retraining the decision model in the policy generation unit.

[0050] In the embodiments disclosed herein, such as Figure 8 As shown, the system also includes: The equalization management module 5 is connected to the battery state estimation module 2 and the adaptive scenario-based energy scheduling module 3, and is used to perform active or passive equalization between battery cells according to the battery state and the optimal energy scheduling strategy. The safety protection module 6 is connected to the data acquisition and processing module 1 and the adaptive scenario-based energy scheduling module 3, and is used to perform emergency protection actions based on the acquired fault signals. The communication and interaction module 7 is connected to the battery state estimation module 2 and the adaptive scenario-based energy scheduling module 3, and is used to realize information interaction between the system and users, the cloud and the power grid.

[0051] It should be noted that, as Figure 9 As shown, the equalization management module 5 includes: an equalization strategy selection unit 5-1, an active equalization circuit control submodule 5-2, a passive equalization triggering submodule 5-3, and an equalization status monitoring unit 5-4; The equalization strategy selection unit 5-1 selects the active or passive equalization mode based on the voltage difference of individual cells and the SOC distribution; the active equalization circuit control submodule 5-2 adjusts the switching state and current of the bidirectional DC-DC converter; the passive equalization trigger submodule 5-3 controls the on / off state of the parallel resistor of the individual cells; and the equalization status monitoring unit 5-4 collects the voltage changes of each individual cell in real time and feeds them back to the strategy selection unit to adjust the equalization intensity.

[0052] It should be noted that, as Figure 10 As shown, the safety protection module 6 includes: a fault detection sensor array 6-1, an overcharge and over-discharge protection submodule 6-2, a temperature protection unit 6-3, and an emergency cut-off actuator 6-4; The fault detection sensor array 6-1 collects abnormal signals such as battery overvoltage, overcurrent, and short circuit; the overcharge and over-discharge protection submodule 6-2 compares the real-time voltage with the preset threshold and triggers the corresponding protection action; the temperature protection unit 6-3 monitors the cell temperature and links the heat dissipation device; the emergency cut-off actuator 6-4 quickly disconnects the battery main circuit when a serious fault occurs to prevent the danger from escalating.

[0053] It should be noted that, as Figure 11 As shown, the communication and interaction module internally includes a local communication interface 7-1, a cloud data transmission unit 7-2, a user interaction interface submodule 7-3, and an instruction parsing submodule 7-4. The local communication interface 7-1 connects the battery management system and the photovoltaic inverter, enabling data exchange between the devices. The cloud data transmission unit 7-2 sends operating data to a remote server via 4G or Wi-Fi. The user interaction interface submodule 7-3 provides status display and instruction input functions. The instruction parsing submodule 7-4 processes user-manually set or grid-dispatched control instructions, converting them into executable operation signals.

[0054] The energy management system proposed in this embodiment has the following advantages: 1. In this invention, the scene recognition unit, through a rule matching library and a clustering analysis submodule, can accurately capture the characteristics of different scenarios such as peak-valley electricity prices, photovoltaic output fluctuations, and emergency backup power for users. Combined with the priority determination submodule, the scenarios are reasonably sorted so that the system can prioritize responding to high-value or urgent needs. The strategy generation unit adopts an improved Q-learning decision model, which can generate optimal charging and discharging and balancing management strategies based on these scenario characteristics. For example, when photovoltaic output is excessive, it can automatically start energy storage charging to absorb excess electricity; when the grid peak electricity price is high, it can release energy storage power to obtain arbitrage profits; and when users experience sudden power outages, it can quickly switch to backup power mode. This not only improves the utilization rate of photovoltaic energy and increases the economic benefits for users, but also ensures the stability of critical electricity demand.

[0055] 2. In this invention, the action execution monitor and real-time data acquisition submodule of the execution feedback unit can continuously track the actual effect after the strategy is executed. Then, the data is integrated by the feedback sample generation unit to form training samples, triggering the model update submodule to optimize the decision model. This closed-loop learning mechanism allows the system to continuously adapt to actual conditions such as battery degradation, changes in user load, and adjustments in power grid policies. The model will automatically adjust the charging and discharging depth to reduce unnecessary losses and extend battery life. At the same time, it can make the strategy more and more in line with actual operating needs, improve the reliability and efficiency of the entire system, and avoid energy waste or failure risks caused by strategy deviation.

[0056] In summary, the energy management system for energy storage batteries proposed in this embodiment solves the problems of poor adaptability and rigid strategies in traditional systems by constructing a closed-loop intelligent scheduling architecture of "perception-decision-feedback". It significantly improves the photovoltaic absorption rate, user economic benefits and battery life, and enhances the system's adaptability to dynamic environments and long-term changes.

[0057] Based on the energy management system for energy storage batteries proposed in the above scheme, this embodiment also proposes an energy management method for energy storage batteries, the method comprising: Real-time acquisition of battery internal status data and external environmental data; Estimate the battery's SOC, SOH, and SOF based on the battery's internal state data; The collected data is identified to determine the current operating scenario and its priority; An improved reinforcement learning decision model is used to generate an optimal energy scheduling strategy based on the scenario, priority, and battery state. Execute the optimal energy scheduling strategy; The execution feedback unit is used to monitor the effect of the strategy execution and generate feedback samples for iterative optimization of the reinforcement learning decision model.

[0058] In summary, the energy management method for energy storage batteries proposed in this embodiment solves the problems of poor adaptability and rigid strategies in traditional systems by constructing a closed-loop intelligent scheduling architecture of "perception-decision-feedback". It significantly improves the photovoltaic absorption rate, user economic benefits and battery life, and enhances the system's adaptability to dynamic environments and long-term changes.

[0059] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0060] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0061] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. An energy management system for an energy storage battery, characterized in that, The system includes: The data acquisition and processing module is used to collect and fuse internal battery status data with external environmental data. The battery state estimation module is connected to the data acquisition and processing module and is used to receive the internal state data of the battery and estimate the battery's state of charge (SOC), state of health (SOH), and state of function (SOF). The adaptive scenario-based energy scheduling module is connected to the data acquisition and processing module and the battery state estimation module respectively, and is used to receive the external environment data and the battery's SOC, SOH, and SOF, and generate the optimal energy scheduling strategy. The execution module is connected to the adaptive scenario-based energy scheduling module and is used to execute the optimal energy scheduling strategy; The adaptive scenario-based energy scheduling module includes: The scene recognition unit is used to extract features and recognize patterns from the external environment data and battery status data to determine the current scene type and its priority. The strategy generation unit has a built-in improved reinforcement learning decision model, which is used to generate the optimal charge and discharge management strategy based on the scene and priority determined by the scene recognition unit and the output of the battery state estimation module. The execution feedback unit is used to monitor the execution effect of the execution module on the strategy and generate feedback samples based on the actual data after execution to optimize the decision model in the strategy generation unit.

2. The management system as described in claim 1, characterized in that, The scene recognition unit includes: The rule matching library pre-stores trigger conditions for multiple preset scenarios; The clustering analysis submodule is used to perform clustering analysis on real-time collected external environment data to identify dynamic sub-patterns in the data. A scene feature extraction unit is used to extract key features for scene determination from the external environment data and battery status data. The priority determination submodule is used to assign or dynamically adjust the priority of the identified current scene based on the matching results of the rule matching library, the output of the clustering analysis submodule, and the key features.

3. The management system as described in claim 2, characterized in that, The preset scenarios include: peak-valley arbitrage scenario, photovoltaic consumption scenario, emergency backup power scenario, and grid auxiliary service scenario; The priority determination submodule is configured to assign the highest priority when an emergency backup power scenario is identified.

4. The management system as described in claim 1, characterized in that, The improved reinforcement learning decision model in the policy generation unit is an improved Q-learning model, and its Q-value update formula is: In the formula, Q(s,a) is the Q value of performing action a in the current state s, α is the learning rate, R is the immediate reward, λ is the battery health cost coefficient, and Csoh is the battery health depletion cost caused by action a. γ is the scene weighting factor, and γ is the discount factor. It is the maximum Q value of all possible actions in the next state s′.

5. The management system as described in claim 1, characterized in that, The adaptive scenario-based energy scheduling module also includes: The AI ​​prediction unit is connected to the data acquisition and processing module and the strategy generation unit, respectively, and is used to predict future user electricity load, photovoltaic output and / or grid electricity price based on historical data and real-time data, and input the prediction results to the strategy generation unit.

6. The management system as described in claim 5, characterized in that, The AI ​​prediction unit incorporates an improved LSTM prediction model, and the forget gate calculation formula for the LSTM prediction model is as follows: In the formula, The output of the forget gate at time t takes a value between 0 and 1, and σ is the Sigmoid activation function. To hide the state of history through the Forgotten Gate and current input The weight matrix, Let be the hidden state vector at time t-1. Let be the input feature vector at time t, and Ws be the weight matrix of the forget gate for the battery's SOC state. Let be the remaining battery charge at time t. This is the bias term for the forget gate.

7. The management system as described in claim 6, characterized in that, The improved LSTM prediction model is trained using a weighted loss function, the formula for which is: In the formula, L is the total model loss, α is the weighting coefficient of the prediction accuracy term, and MAE is the mean absolute error. The calculation formula is as follows: Where: n is the total number of samples involved in the calculation, and i is the sample index, ranging from 1 to n, representing the i-th specific sample. for , The true value of the i-th sample, |...| is used to convert the difference between the predicted value and the true value into a non-negative number, avoiding the cancellation of positive and negative errors. For the predicted electricity load vector, Let be the actual electrical load vector, and β be the weighting coefficient of the SOC constraint term. For the predicted battery SOC value, This represents the user's desired battery SOC target value.

8. The management system as described in claim 1, characterized in that, The execution feedback unit includes: An action execution monitor is used to track and record the actual execution of the strategy by the execution module; The real-time data acquisition submodule is used to collect system status data at a preset frequency after the strategy is executed. The feedback sample generation unit is used to integrate the state data before the action is executed, the action information after the action is executed, and the state data after the action is executed into a standardized feedback sample. The model update triggering submodule is used to initiate the update of the decision model in the strategy generation unit based on preset conditions and the generated feedback samples.

9. The management system as described in claim 1, characterized in that, The system also includes: The equalization management module, connected to the battery state estimation module and the adaptive scenario-based energy scheduling module, is used to perform active or passive equalization between battery cells according to the battery state and the optimal energy scheduling strategy. The safety protection module is connected to the data acquisition and processing module and the adaptive scenario-based energy scheduling module, and is used to perform emergency protection actions based on the acquired fault signals. The communication and interaction module is connected to the battery state estimation module and the adaptive scenario-based energy scheduling module, and is used to realize information interaction between the system and users, the cloud and the power grid.

10. A method for managing the energy storage battery based on the energy storage battery energy management system according to any one of claims 1-9, characterized in that, The method includes: Real-time acquisition of battery internal status data and external environmental data; Estimate the battery's SOC, SOH, and SOF based on the battery's internal state data; The collected data is identified to determine the current operating scenario and its priority; An improved reinforcement learning decision model is used to generate an optimal energy scheduling strategy based on the scenario, priority, and battery state. Execute the optimal energy scheduling strategy; The execution feedback unit is used to monitor the effect of the strategy execution and generate feedback samples for iterative optimization of the reinforcement learning decision model.

Images